Arabic - Multiple questions

[Kentoseth]

Hello/Bonjour.

I am experimenting with https://github.com/inception-project/inception/ at the moment, but there are some drawbacks/quirks due to the nature of the complex morphology and grammar in Arabic. I have a few questions:

• Can I create my own PoS and Dependency layers?
• Can my PoS layer be at the letter level? Example here: https://corpus.quran.com/treebank.jsp?chapter=1&verse=1&token=1 (the first word 'bism')
• Can I setup a Phrase layer on top of a Dependency layer? (or stack the phrase layer on top of the other PoS layer)? Example here: https://corpus.quran.com/treebank.jsp?chapter=112&verse=1 (the NS = Nominal Sentence which equates to: PRON + PN + N)
• Can I make a Dependency layer point to any PoS layer? Example here: https://corpus.quran.com/treebank.jsp?chapter=112&verse=1 (مفعول به = dependency layer pointing back to the verb 'qul')

Thanks.

[jheinecke]

Hi, ConlluEditor is primarily made for editing CoNLL-U files of the Universal Dependencies (UD) project. Multiple syntax layers are not possible in the CoNLL-U format. However you can use a different POS tagset than the one provided by UD. But a POS as assigned to a token, not to a character.

So for your example bismi you have to create a MultiTokenWord (ConlluEditor can do this) bi + smi in order to assign differnet POS to the preposition biand the noun smi. So your first example in CoNLL-U would look like

1-2 bi'smi  _   _   _   _   _   _   _   _
1   bi  bi  ADP _   _   2   case    _   _   
2   'smi    'sm NOUN    _   Gender=Masc 0   root    _   _
3   l-lahi  l-lah   PROPN   _   Case=Gen    2   nmod    _   _
4   l-raḥmāni   l-raḥmān    ADJ _   _   3   amod    _   _
5   l-raḥīmi    l-raḥīm ADJ _   _   3   amod    _   _

(Have alook in the Arabic Treebank for better examples how Arabic can be dealt with within UD)